Large-Scale Deep Learning-Enabled Infodemiological Analysis of Substance Use Patterns on Social Media: Insights From...

Area of Science:

• Digital epidemiology
• Public health surveillance
• Computational social science

Background:

• COVID-19 pandemic exacerbated mental health and substance use (SU) challenges.
• Increased societal stress led to higher drug reliance and overdose rates.
• Social media engagement surged, offering insights into public health trends.

Purpose of the Study:

• Analyze SU patterns using large-scale social media data during the COVID-19 pandemic.
• Examine prepandemic, pandemic, and postpandemic periods for baseline and consequence analysis.
• Provide a comprehensive understanding of SU trends across various drug types and underlying themes.

Main Methods:

• Utilized a deep learning model (RoBERTa) to analyze 1.13 billion Twitter posts (2019-2021).
• Employed a human-in-the-loop strategy to enhance model performance and data annotation.
• Applied statistical techniques (trend analysis, clustering, topic modeling) and integrated a real-time monitoring system.

Main Results:

• Identified 9 million SU-related posts between 2019 and 2021.
• Observed a significant 21% increase in SU posts within 3 days of the pandemic declaration in 2020.
• Alcohol, cannabinoids, and prescription medications saw increased discussion, linked to COVID-19, mental health, and economic stress.

Conclusions:

• Social media data enables real-time detection of SU trends during global crises.
• Mental health and economic stress are key drivers of SU spikes, especially for alcohol and prescription drugs.
• Data-driven insights support proactive public health interventions for vulnerable populations.